Engineering notes on building
dependable AI-enabled systems.
Retrieval design · 12 MIN READ
Start retrieval design with the question and the data: exact SQL relationships, permission and version boundaries, and where semantic search belongs. Includes a runnable example.
Evaluation and testing · 11 MIN READ
Test LLM-generated artifacts without requiring identical wording: check structure, policy correctness, coverage, and unsupported claims with a runnable evaluation example.
Execution reliability · 13 MIN READ
An LLM job saves 13 of 20 rules, then crashes. Here is how I design checkpoints, retries, idempotency, and worker ownership so it can resume safely.
Change impact analysis · 11 MIN READ
A practical guide to change impact analysis in LLM workflows: classify specification changes, trace dependencies, and plan selective regeneration.
Built from real engineering problems, with original examples you can run.
For the projects behind the writing, explore my work.
I use Google Analytics to understand which articles and projects are useful. Allow optional analytics cookies? Google’s privacy policy